← All case studies
Document AI · Logistics

Intelligent document processing

An OCR/IDP pipeline that reads, understands and routes millions of shipping documents at near-perfect accuracy.

Client
A global logistics operator
Sector
Logistics & Shipping
Region
Global
Engagement
Dedicated team
Timeline
Phased by document type
99.4%
Field accuracy
12M+
Documents processed
Hours → sec
Per document
01

The brief

A logistics operation drowned in paper — bills of lading, customs forms, invoices, proofs of delivery — arriving as scans and phone photos in countless layouts. Teams keyed them by hand, which was slow, error-prone, and impossible to scale to the millions of documents flowing through the network.

They needed a pipeline that could read almost anything, extract and validate the fields that mattered, route each document correctly, and escalate only genuine exceptions — at a scale and accuracy people simply couldn't match.

What the client asked for
  • Read and route millions of shipping documents automatically.
  • Handle scans, photos and many layouts — including handwriting.
  • Reach near-perfect field accuracy on validated documents.
  • Validate against business and customs rules; escalate only uncertain cases.
  • Maintain a verifiable audit trail at document scale.
  • Scale to peak volumes without adding headcount.
02

Our AI-native approach

We built an IDP/ICR pipeline that treats each document as a sequence of scored decisions — classify, extract, validate, route — each carrying a confidence value. High-confidence documents pass straight through; anything uncertain goes to a focused human review with the model's best guess pre-filled.

We began by mining the real document corpus to understand the true distribution of layouts, languages and edge cases before writing a line of extraction logic.

03

What we built

Any-format intake

Scans, photos and digital files are normalised, de-skewed and quality-scored on entry.

Classification & routing

Each document is identified by type and sent to the right workflow.

OCR & ICR extraction

Printed and handwritten fields are read, with confidence on every value.

Rule validation

Extracted data is checked against business and customs rules automatically.

Exception handling

Only low-confidence documents reach a human, pre-filled with the model's proposal.

Audit trail at scale

Every field carries its source and confidence across millions of documents.

04

How we built it

Discovery started with the corpus itself, so the pipeline was designed for reality rather than an idealised sample. We made confidence a first-class concept early, because it is what lets the system route work safely between automation and human review.

We delivered by document type in phases, each gated on accuracy against a human-labelled set before going live. Reviewer corrections fed back into training from day one, so accuracy compounded as volume grew.

05

How it works

1

Ingest

Documents are normalised and quality-scored.

2

Classify

Type and layout are identified.

3

Extract

OCR/NLP pulls structured fields with confidence.

4

Validate

Rules and cross-checks confirm or flag each value.

5

Route

Clean documents pass through; uncertain ones go to review.

The intelligence layer

The extraction stack pairs OCR with NLP classification and named-entity models tuned to the document set. Confidence is first-class: it decides routing, so the system gets faster as the models learn, and reviewers spend their time only on genuine ambiguity.

Validated corrections feed back into training, so at twelve million documents accuracy compounds rather than plateaus.

06

The impact

99.4%
Field accuracy
12M+
Documents processed
Hours → sec
Per document

Manual document handling fell dramatically while throughput rose.

Field accuracy held near-perfect across millions of documents.

Peak volumes were absorbed without temporary staffing.

Reviewers shifted from data entry to genuine exception handling.

07

Technology stack

Document AI
OCRICRNLPNamed-entity recognition
Platform
Azure AIFastAPIPython
Workflow
Confidence routingHuman-in-the-loop
Governance
Field-level auditRule validation

Building something similar?

If this maps to a problem you're facing, tell us what you're building. We'll show you how we'd engineer it — and come back within one business day.